Wafer Map Defect Pattern Recognition Using Rotation-Invariant Features

In semiconductor manufacturing, the patterns on the wafer map provide important information for engineers to identify the root causes of production problems. The detection and recognition of wafer map patterns is thus an important issue in semiconductor industry. Automatic techniques are required to cut down on cost and to improve accuracy. In this study, we propose an approach to recognize patterns in the wafer maps which uses the extracted features based on the proposed weight masks. The proposed masks contain three types, namely, polar masks, line masks and arc masks. Polar masks aim to extract features of concentric patterns, while line and arc masks are designed to mainly deal with eccentric patterns like scratches. These masks can be applied to extract rotation-invariant features for the classification of the defect patterns. To demonstrate the effectiveness of our model, we apply the method to a real-world wafer map dataset. Comparisons with alternative methods show superiority of our method in the task of wafer map defect pattern recognition.